Causality Is Not a Flat List

A layered map of causal classifications, and a repair of “horizontal” and “vertical” causality

Terminology and taxonomy note · causation, causal structure, dependence, explanation, and conceptual hygiene

Andraž Đurič

Version 1 · August 2026

Abstract

The literature surveyed for this note does not supply a single canonical list into which everything called a “type of causality” cleanly fits. It contains competing accounts of what causation is, properties of causal relations, positions in causal structures, roles played by causal factors, levels of generality, methods of causal inference, and nearby explanatory relations that may not be causal at all. Flattening these into one list produces category errors.

This note proposes a layered classification map. It does not define causation or replace existing theories. It first asks which classificatory question is being answered, then permits a causal claim to receive several compatible descriptors. A relation may be type-level, probabilistic, diachronic, mediated, sustaining, and contributory at once. These labels are not competitors.

The map exposes a defect in the pair “horizontal causality” and “vertical causality.” “Horizontal” often has a recoverable target: causal succession across time, better called diachronic causation or, in ordinary language, causal propagation through time. “Vertical” has no comparably stable target. It has been used for sustaining, simultaneous, constitutive, cross-level, grounding, metacausal, and theological relations. In the familiar example of a couch holding up a body, the load-bearing feature is ongoing maintenance. The precise term is sustaining causation, not spatial causality and not automatically synchronic causation.

The proposal is methodological: name the relata, identify the question, and then name the relation. Diagrammatic orientation should not substitute for causal structure.

1. The phrase “types of causality” hides several questions

Ask for the types of causality and the answers arrive at different logical levels.

One list gives counterfactual, probabilistic, interventionist, mechanistic, process, or powers-based causation. Another gives direct, indirect, necessary, sufficient, upward, downward, deterministic, probabilistic, simultaneous, backward, preventive, or sustaining causation. A third adds experiments, structural causal models, Bayesian networks, or Granger methods. A fourth includes constitution, realization, grounding, function, or explanation.

These are not entries in one partition.

A counterfactual theory is an account of what makes causal claims true. Indirectness describes a pathway. Necessity and sufficiency describe a factor's modal role relative to background conditions. Downwardness locates a relation across an organizational hierarchy. A structural causal model represents causal assumptions and supports inferences. Constitution may connect parts and wholes without being causation at all.

Before asking what type of causality something is, ask what kind of classification is being attempted.

The absence of a canonical flat list is therefore not surprising. The candidate entries answer different questions.

Search-status firewall. The claim is not that nobody has classified causation, nor that no broader taxonomy exists. Many taxonomies, handbooks, formal frameworks, and pluralist proposals do. The bounded claim is that the sources surveyed for this note did not yield one canonical, cross-disciplinary, finite list accepted as the taxonomy of causal types. Failure to find such a list is not proof that none has ever been proposed.

2. What the literature already supplies

Accounts of causation

Major traditions analyze causation through regularities or laws, counterfactual dependence, probability change, interventions, mechanisms, physical processes, or powers and dispositions. Lewis's counterfactual account, Suppes's probabilistic theory, Woodward's interventionism, and mechanistic accounts in philosophy of science are prominent examples.

These accounts can conflict, overlap, or target different work done by causal language. Hall's distinction between dependence and production is especially important: an effect may depend counterfactually on a cause without a physical process being transmitted from one to the other, while production can survive cases in which simple counterfactual dependence fails. Causal pluralists take such fractures seriously, although pluralism itself remains disputed.

This is a debate about the concept or worldly relation of causation. It is not yet a classification of a particular causal pathway.

Descriptors of causal claims and relations

The literature also distinguishes:

token or singular claims from type-level or general claims;
deterministic from probabilistic causal relations or models;
direct, indirect, mediated, total, and path-specific effects;
necessary, sufficient, component, and contributing conditions, including Mackie's INUS conditions;
productive, preventive, omission-based, and double-preventive cases;
diachronic, allegedly simultaneous, backward, and feedback-involving temporal structures;
one-hit and sustaining causation;
same-level, upward, downward, and cross-level causal descriptions.

Most of these distinctions are not mutually exclusive. Smoking can be described as a type-level, probabilistic, diachronic, multiply mediated, contributory cause of lung cancer. Each term answers a different question.

Representations and warrants

Experiments, interventions, causal graphs, structural equations, potential outcomes, time-series methods, and mechanistic evidence concern how causal claims are represented, identified, estimated, or warranted. They may embody substantive accounts of causation, but they should not be inserted into a flat list beside “indirect cause” as though both classified the same feature.

Pearl's structural causal models, for example, unify graphical, probabilistic, interventionist, and counterfactual machinery. Within that framework, direct and indirect effects are formally defined relative to a model and its chosen variables. “Direct” therefore does not mean metaphysically unmediated in every possible description. It means unmediated by variables represented as intermediaries in the model.

Relations near causation

Constitution, realization, grounding, mathematical dependence, functional explanation, and evidential relevance can explain without thereby being causal relations.

This boundary matters most in part–whole and cross-level cases. Craver and Bechtel argue that many apparent top-down and bottom-up causes are better analyzed as mechanistically mediated effects: intralevel causal relations combined with interlevel constitutive relations. Whether that analysis covers every case remains disputed. Its decisive lesson here is narrower:

Different explanatory dependence is not automatically a different kind of causation.

3. Why a flat taxonomy fails

A sound taxonomy partitions items according to a common question. “Counterfactual,” “mediated,” and “downward” do not compete for one slot because they classify different features.

Flattening them causes three failures.

Category mixing

A theory, a relation-profile, a structural position, and an inferential method are treated as equivalent kinds of item.

False exclusivity

Readers are invited to choose one label where several can be true together. A probabilistic cause does not cease to be indirect because it is probabilistic.

Causal inflation

Any relation that explains, constitutes, realizes, supports, correlates with, or grounds another is admitted as a form of causation. The word expands until it stops discriminating.

Causal vocabulary is faceted. A flat list deletes the facets and then mistakes the resulting ambiguity for unity.

4. A layered map of causal classifications

The repair is not one longer list. It is a map organized by question.

Layer	Question	Representative answers
Account	What makes this relation causal?	Regularity, counterfactual dependence, probability change, intervention, mechanism, process, powers; dependence and production
Claim scale	What is the claim about?	Token or singular; type-level or general; population-level; actual causation
Modal profile	How strongly does the cause constrain the effect?	Deterministic or probabilistic; necessary, sufficient, INUS, component, contributing
Temporal and dynamical profile	How are cause and effect organized through time?	Diachronic, allegedly simultaneous, backward, feedback; one-hit or sustaining
Pathway and topology	Through what represented route does influence run?	Direct, indirect, mediated, total, path-specific; common cause; chain; feedback loop
Operation and structural location	What does the factor do, and where is it represented?	Production, prevention, omission, double prevention; same-level, upward, downward, cross-level
Warrant and representation	How is the claim modeled, detected, or supported?	Intervention, experiment, structural causal model, potential outcomes, mechanistic evidence, temporal prediction
Boundary	Is the relation causal at all?	Causation versus constitution, realization, grounding, correlation, evidential relevance, function, or other explanation

This is a layered classification map, not a strict Cartesian coordinate system.

The layers are not guaranteed to be exhaustive, mutually independent, or neutral between every theory of causation. “Deterministic” may characterize a model, a law-governed system, or a relation depending on context. “Direct” changes with variable choice. “Downward” depends on a defensible account of levels. “Prevention” may be analyzed differently by dependence and production theories.

The map's function is prior to settlement of those disputes. It prevents an answer to one question from being mistaken for an answer to another.

5. How to classify a causal claim

Begin with the claim, not the vocabulary.

Name the relata. What exactly are the candidate cause and effect: events, states, processes, variable values, substances, omissions, or something else?
Name the model and grain. Which variables, background conditions, organizational level, time-scale, and system boundary are being used?
State the causal commitment. Is the claim about counterfactual difference-making, intervention, production, mechanism, probability change, or another relation?
Add only earned descriptors. Temporal order, pathway, strength, causal role, operation, level, and claim scale should be specified only when they do explanatory work.
Check the boundary. Is a constitutive, grounding, predictive, or evidential relation being mislabeled as causal?

The resulting description may look like this:

Relative to model M and background B, C is a type-level, probabilistic, diachronic, mediated, contributing cause of E.

This grammar does not prove the causal claim. It makes the claim inspectable. Evidence is still required for every filled slot.

6. Repairing “horizontal causality”

“Horizontal causality” usually presents causes as a line: one event produces another, which produces another, across time.

The metaphor is readable, but the page orientation carries no causal information. Rotate the diagram and the relation remains unchanged.

If the intended contrast is earlier-to-later succession, the better terms are:

ordinary language: causal propagation through time;
accessible formal language: temporal causal propagation;
technical language: diachronic causation.

“Diachronic” is more exact than “temporal” because temporal organization also includes disputes about simultaneous causation, backward causation, and feedback. It says that the relevant relata occupy different times.

The term does not imply determinism. A diachronic causal relation may be modeled deterministically or probabilistically. Nor is the butterfly effect another name for it. The butterfly effect is sensitive dependence on initial conditions in certain dynamical systems; it is one possible pattern within temporal evolution, not a synonym for causal succession.

7. Why “vertical causality” cannot be renamed wholesale

“Vertical causality” appears to complete a pair. That symmetry is misleading.

Across philosophical traditions and contemporary discussions, the phrase can indicate:

an ongoing cause maintaining an effect;
a supposedly simultaneous cause and effect;
a part–whole or constitutive relation;
upward, downward, or cross-level influence;
the conditions explaining why one causal relation obtains;
grounding, formal dependence, or realization;
atemporal, primary, or theological causation.

These are not one relation awaiting a better name. They are several relations compressed by one diagrammatic metaphor.

The couch case

Consider a body resting on a couch. The couch does not merely initiate the body's position and then become irrelevant. Its support must continue. Remove it and, absent another support, the body falls.

The decisive contrast is therefore between a transient cause that leaves an irreversible result and a continuing cause required to maintain a reversible state. Ross and Woodward call the latter sustaining causation.

For the couch example, replace “vertical causality” with “sustaining causation” or, in ordinary language, “ongoing causal support.”

Calling the case “synchronic causation” would make an additional claim. Genuine simultaneity between distinct cause and effect is philosophically disputed. At a coarse time-scale, couch support and bodily position can be modeled as contemporaneous. At a finer physical scale, the maintained state is realized through continuous interactions with temporal structure. “Sustaining” captures what matters without settling the simultaneity dispute.

Other cases require other terms

If parts and organization make up a whole, use constitutive dependence.
If one organizational level influences another, use upward, downward, or cross-level causal influence, while checking whether the case is instead a causal–constitutive hybrid.
If distinct cause and effect are claimed to coexist in time, use simultaneous causation and defend that claim.
If one fact obtains in virtue of another, use grounding where that technical relation is intended.
If the question concerns why C causes E, use causal mechanism or metacausal explanation, depending on the claim.

“Spatial causality” is not a general solution. Diachronic causation can cross space, while sustaining, simultaneous, and constitutive relations can be spatially local or distributed. Space specifies where relata are, not what relation connects them.

8. The terminological proposal

Intended relation	Ordinary term	Technical term
Earlier states or events affect later ones	Causal propagation through time	Diachronic causation
A continuing cause maintains a reversible state	Ongoing causal support	Sustaining causation
Parts and organization make up a whole	Part–whole dependence	Constitutive dependence
A causal relation crosses organizational levels	Cross-level influence	Upward or downward causal influence, if warranted
Distinct cause and effect occur at the same time	Simultaneous causation	Synchronic causation
One fact obtains in virtue of another	In-virtue-of dependence	Grounding

The proposal is asymmetrical:

“Horizontal causality” usually has a recoverable replacement. “Vertical causality” usually requires disambiguation.

This asymmetry is the result. The original pair makes two heterogeneous collections appear to be two unified causal kinds.

9. Limits and revision conditions

This map should be revised if it fails to keep the distinctions it was built to protect.

A canonical taxonomy is identified

If a cross-disciplinary taxonomy already separates these classificatory layers more completely and has stable uptake, the novelty claim contracts to exposition or terminological application.

The layers do not improve classification

If trained readers classify causal claims no more consistently with the layered map than with an ordinary flat list, the map has not earned its added structure.

The map hides theory-dependence

If it assigns apparently neutral coordinates that covertly presuppose one disputed account of causation, those coordinates must be marked as framework-relative or removed.

“Vertical causality” proves stable

If systematic corpus work shows a dominant, precise, and cross-contextual use that the present survey missed, the recommendation to disambiguate must narrow to the remaining uses.

The boundary is misdrawn

If constitution, grounding, or realization is shown to be causal in the relevant sense, the map must change. It should not preserve a distinction merely because a literature currently uses it.

What terminology can and cannot do. Better labels can expose commitments and prevent category errors. They cannot settle whether causation is fundamentally one relation or many, whether simultaneous or backward causation exists, whether downward causation is irreducible, or which causal model is correct.

10. The architecture, compressed

The literature does not contain one kind of causal classification.

It contains debates over what causation is, descriptors of causal relations, roles within sufficient structures, locations in causal graphs and organizational levels, methods of inference, and relations adjacent to causation.

A flat list mixes them.

The repair is to classify the classification before classifying the cause.

Name the relata. Name the model and grain. Identify the classificatory question. Then name the relation.

Under that discipline, “horizontal causality” usually resolves into diachronic causation.

“Vertical causality” does not resolve into one thing. In the couch case it becomes sustaining causation. In other cases it may become constitution, cross-level influence, simultaneous causation, grounding, or metacausal explanation.

The gain is not verbal elegance.

It is the removal of false unity.

Standing of this essay

The distinctions surveyed here are established in their respective literatures. The layered arrangement is a synthesis proposed in this note.

The strongest claim is diagnostic: lists of “types of causality” often mix accounts, relation-properties, structural roles, methods, and noncausal dependencies. These cannot form one clean partition because they answer different questions.

The map is intentionally non-exhaustive. It is designed to preserve major distinctions and prevent category mixing, not to enumerate every causal term used in philosophy, science, statistics, law, biology, or physics.

The claim that no canonical cross-disciplinary finite taxonomy was found reports the result of this survey. It is not a proof of universal absence.

The repair of “horizontal causality” is relatively secure when the phrase denotes earlier-to-later succession. The analysis of “vertical causality” is more context-sensitive. Its recommended replacement depends on the relata and relation in the case at hand.

The classification of the couch case as sustaining causation is stronger than its classification as synchronic causation. The former identifies the continuing-maintenance profile; the latter adds a disputed thesis about temporal coincidence.

The proposal should be judged by whether it makes causal claims easier to specify, compare, contest, and correct.

References

Craver, C. F., & Bechtel, W. (2007). “Top-Down Causation Without Top-Down Causes.” Biology & Philosophy, 22, 547–563. — argues that many interlevel causal descriptions are hybrids of intralevel causation and interlevel constitution. Verified
Dinh, P. N., & Danks, D. (2021). “Causal Pluralism in Philosophy: Empirical Challenges and Alternative Proposals.” Philosophy of Science, 88(5), 761–772. — challenges a prominent empirical argument for causal pluralism and shows that pluralism remains contested. Verified
Godfrey-Smith, P. (2010). “Causal Pluralism.” In H. Beebee, C. Hitchcock, & P. Menzies (eds.), The Oxford Handbook of Causation, 326–337. Oxford University Press. — frames causal pluralism as the view that causal language masks more than one kind of relation or connection. Verified
Gurtler, G. M. (2008). “Plotinus on the Soul's Omnipresence in Body.” International Journal of the Platonic Tradition, 2(2), 113–127. — exemplifies a metaphysical use contrasting horizontal causality of change with vertical causality of participation. Verified
Hall, N. (2004). “Two Concepts of Causation.” In J. Collins, N. Hall, & L. A. Paul (eds.), Causation and Counterfactuals, 225–276. MIT Press. — develops the distinction between dependence and production. Verified
Huemer, M., & Kovitz, B. (2003). “Causation as Simultaneous and Continuous.” The Philosophical Quarterly, 53(213), 556–565. — defends the controversial thesis that actual causes are simultaneous with their direct effects. Verified
Lewis, D. (1973). “Causation.” The Journal of Philosophy, 70(17), 556–567. — foundational counterfactual analysis of singular causation. Verified
Machamer, P., Darden, L., & Craver, C. F. (2000). “Thinking About Mechanisms.” Philosophy of Science, 67(1), 1–25. — analyzes mechanisms through organized entities and activities productive of regular change. Verified
Mackie, J. L. (1965). “Causes and Conditions.” American Philosophical Quarterly, 2(4), 245–264. — introduces the INUS analysis of causal conditions. Verified
Pearl, J. (2001). “Direct and Indirect Effects.” In Proceedings of the Seventeenth Conference on Uncertainty in Artificial Intelligence, 411–420. — formalizes direct and indirect effects within structural causal models. Verified
Pearl, J. (2009). Causality: Models, Reasoning, and Inference (2nd ed.). Cambridge University Press. — integrates graphical, structural, probabilistic, interventionist, and counterfactual causal analysis. Verified
Ross, L. N., & Woodward, J. F. (2022). “Irreversible (One-Hit) and Reversible (Sustaining) Causation.” Philosophy of Science, 89(5), 889–898. — distinguishes transient irreversible causes from causes that must continue to maintain reversible effects. Verified
Suppes, P. (1970). A Probabilistic Theory of Causality. North-Holland. — systematic probabilistic account of causal relations. Verified
Woodward, J. (2003). Making Things Happen: A Theory of Causal Explanation. Oxford University Press. — develops an interventionist account centered on invariant change-relations and manipulation. Verified

Verification key: “Verified” means the bibliographic identity and the attributed use were checked against the linked paper, publisher record, or authoritative repository during preparation of this version. The sources establish the component distinctions; they do not jointly entail the layered synthesis proposed here.

Internal lineage

This note applies the Epistemic Forge instruction to name the relata before assigning a relation. It extends the method from objectivity and semantic localization to causal vocabulary.

The contribution claimed here is synthetic and terminological: separating causal classifications by the question they answer; treating multi-label description as legitimate without pretending the layers are perfectly orthogonal; and deriving an asymmetrical repair of “horizontal” and “vertical” causality from that map.

Search This Blog

Epistemic Forge